Network Meta-Analysis of Survival Outcomes With Non-Proportional Hazards Using Flexible M-Splines

Area of Science:

• Biostatistics
• Health Economics
• Pharmacometrics

Background:

• Network meta-analysis (NMA) is crucial for comparing multiple treatments in healthcare decision-making.
• Standard NMA models often assume proportional hazards, which may not hold for novel therapies like immunotherapies.
• Non-proportional hazards necessitate flexible modeling approaches beyond standard parametric methods.

Purpose of the Study:

• To propose a novel, flexible NMA model capable of handling non-proportional hazards in time-to-event outcomes.
• To develop a robust Bayesian framework using M-splines and weighted random walk priors for accurate hazard function modeling.
• To implement the proposed methods in an accessible R package (multinma) for broader application.

Main Methods:

• Utilized M-splines on the baseline hazard function for enhanced flexibility.
• Introduced a novel weighted random walk prior distribution for shrinkage and invariance.
• Modeled non-proportional hazards by stratifying by treatment or incorporating treatment effects on spline coefficients.
• Developed an R package, multinma, supporting aggregate and individual participant data.

Main Results:

• The proposed M-spline NMA model effectively accommodates non-proportional hazards.
• The weighted random walk prior prevents overfitting and is robust to parameter choices.
• The multinma package provides a user-friendly implementation for complex NMA.
• Demonstrated application in a NMA of non-small cell lung cancer treatments for progression-free survival.

Conclusions:

• The developed M-spline NMA model offers a flexible and robust solution for time-to-event data with non-proportional hazards.
• This approach enhances the reliability of treatment effect estimates for novel therapies.
• The multinma package facilitates the application of advanced NMA techniques in clinical and health economic evaluations.